To reduce the risk of transfer, lifting and repositioning (TLR)-related repeated injuries from patient handling, an ergonomic injury prevention program was implemented for the healthcare workers in the intervention group. However, no intervention program was implemented for the control group. Both the intervention and control groups were contained within three hospital sizes: large, medium and small. The goal of this study was to investigate the effectiveness of the injury prevention program using the multiplicative Cox and additive hazards models. Based on our analysis, the results indicate that the TLR intervention program was effective and sustained for healthcare workers by reducing repeated injuries induced by patient handling. Therefore, the risks of patient handling-related repeated injuries among healthcare workers can be lowered by implementing a multi-factor TLR intervention program with the right equipment and training.
The TLR intervention program was implemented for the healthcare workers in the intervention group as compared to the control group. The multivariate Cox model showed that the intervention group had 27% fewer repeated injuries than the control group, which indicates the effectiveness of the TLR intervention program. The intervention group also showed protection from repeated TLR injures by the L-Y additive hazards model. The intervention group had 0.002 fewer hazards for repeated injuries than the control group, which supports the result of the Cox model. On the other hand, nurses and nursing aides had the most repeated injuries by occupation; the Cox model showed a 72% higher hazard of repeated injuries than other occupations. According to the L-Y model, nurses and nursing aides had a 0.0024 higher risk of repeated injuries than other occupations. Nurses and nursing aides are directly involved in patient handling. Although the TLR intervention program was implemented for the intervention group, they still had a higher risk of repeated injuries regarding TLR, which showed significance in the model. Among all of the healthcare workers, body parts were the most significant risk factor for repeated injuries. The Cox model indicated that the back, neck and shoulders had a 115% increased risk of repeated injuries as compared to other body parts. The L-Y additive hazards model also showed that the back, neck and shoulder had a 0.0038 increased risk of repeated injures than the others body parts. Both the Cox and L-Y models, as well as Aalen’s additive
hazards model, showed that the TLR intervention program had a significant impact on reducing repeated injures among healthcare workers.
The Cox model is the most widely used model for the analysis of survival data in clinical research. However, the proportional hazard assumption may not always be satisfied in the data. In such cases, there are various solutions to consider; for example, inclusion of a time-dependent covariate. While the coefficients in the Cox model act in a multiplicative way on unknown baseline hazards, coefficients in the additive hazards models act in an additive way on unknown baseline hazards. Because the coefficients act in different ways in the multiplicative and additive hazards models, it is very difficult to compare them directly. In this thesis, the multiplicative and additive hazards models similarly identified the significant covariates of the repeated injuries among healthcare workers. However, the different models interpreted the coefficients in different ways. The association between the covariates and the time to repeated injuries in the additive hazards models was explained in terms of the risk difference or excess risk rather than the risk ratio. However, if one would like to estimate the cumulative hazard of an event for more extreme values, the additive and the Cox hazards models estimates are remarkably different. By using the time varying covariates effect, this can be settled on by which are taken into account by the additive hazards model but not by a multiplicative Cox hazards model. Moreover, when using a more compromised covariate profile, the multiplicative model gives a higher estimate than the additive model, probably because of the multiplicative effect of fixed covariates on baseline function (21).
In this thesis, administrative data was used that had been supplied by two Health Regions. The data acquisition, injury classification criteria, and data extraction process could not be controlled or evaluated. The lack of information on the subjects’ demographic and injury characteristics and the total number of direct care workers employed at each site weakened the study. Another drawback of the data was that there was no identification number for the control group. Thus, in case of identified the repeated injury there would be personal selection bias. The additive models have some limitations. Aalen’s model may provide more in-depth information on the effect of a prognostic factor over time. However, one has to visualize all covariates’ effects over time, and a
simple interpretation of the effects is not possible, which makes Aalen’s model less appealing in real applications than other models. A theoretical limitation of the Lin and Ying additive hazards model is that the linear predictors in the model constrain to be positive (19). A very practical limitation of the additive hazards models is the availability of computer programs. For the Cox hazards model, various statistical software packages are available, and it is easy to fit the models. However, for the additive hazards model, any standard procedure is limited to SAS, R, and Stat. Few macros are available for the analysis of goodness of fit (22, 84). Because the different macros are not used globally, it will be difficult to make a real comparison.
In many applications, the additive hazards models are plausible and often attractive in epidemiologic applications, where the baseline hazards is taken to be the baseline mortality of the population and the coefficient measures the excess risk of the patients under study. As an example, in a study of diabetic patients (85), if the measured covariates predict the severity of disease and its downstream mortality/morbidity, but have no impact on independent causes of death, such as malignancy, then the multiplicative hazards model might not be appropriate. In such cases, the additive hazards model may be better for patients with more severe clinical profiles, which is relevant to the development of patient management and care (86). The risk difference can be more important than the risk ratio in understanding an association between a risk factor and disease occurrence (19). The results of this study are also consistent with another published study (22).